1 /* 2 * Virtio Balloon Device 3 * 4 * Copyright IBM, Corp. 2008 5 * Copyright (C) 2011 Red Hat, Inc. 6 * Copyright (C) 2011 Amit Shah <amit.shah@redhat.com> 7 * 8 * Authors: 9 * Anthony Liguori <aliguori@us.ibm.com> 10 * 11 * This work is licensed under the terms of the GNU GPL, version 2. See 12 * the COPYING file in the top-level directory. 13 * 14 */ 15 16 #include "qemu/osdep.h" 17 #include "qemu/iov.h" 18 #include "qemu/timer.h" 19 #include "qemu-common.h" 20 #include "hw/virtio/virtio.h" 21 #include "hw/mem/pc-dimm.h" 22 #include "sysemu/balloon.h" 23 #include "hw/virtio/virtio-balloon.h" 24 #include "exec/address-spaces.h" 25 #include "qapi/error.h" 26 #include "qapi/qapi-events-misc.h" 27 #include "qapi/visitor.h" 28 #include "trace.h" 29 #include "qemu/error-report.h" 30 31 #include "hw/virtio/virtio-bus.h" 32 #include "hw/virtio/virtio-access.h" 33 34 #define BALLOON_PAGE_SIZE (1 << VIRTIO_BALLOON_PFN_SHIFT) 35 36 static void balloon_page(void *addr, int deflate) 37 { 38 if (!qemu_balloon_is_inhibited() && !deflate) { 39 qemu_madvise(addr, BALLOON_PAGE_SIZE, QEMU_MADV_DONTNEED); 40 } 41 } 42 43 static const char *balloon_stat_names[] = { 44 [VIRTIO_BALLOON_S_SWAP_IN] = "stat-swap-in", 45 [VIRTIO_BALLOON_S_SWAP_OUT] = "stat-swap-out", 46 [VIRTIO_BALLOON_S_MAJFLT] = "stat-major-faults", 47 [VIRTIO_BALLOON_S_MINFLT] = "stat-minor-faults", 48 [VIRTIO_BALLOON_S_MEMFREE] = "stat-free-memory", 49 [VIRTIO_BALLOON_S_MEMTOT] = "stat-total-memory", 50 [VIRTIO_BALLOON_S_AVAIL] = "stat-available-memory", 51 [VIRTIO_BALLOON_S_CACHES] = "stat-disk-caches", 52 [VIRTIO_BALLOON_S_HTLB_PGALLOC] = "stat-htlb-pgalloc", 53 [VIRTIO_BALLOON_S_HTLB_PGFAIL] = "stat-htlb-pgfail", 54 [VIRTIO_BALLOON_S_NR] = NULL 55 }; 56 57 /* 58 * reset_stats - Mark all items in the stats array as unset 59 * 60 * This function needs to be called at device initialization and before 61 * updating to a set of newly-generated stats. This will ensure that no 62 * stale values stick around in case the guest reports a subset of the supported 63 * statistics. 64 */ 65 static inline void reset_stats(VirtIOBalloon *dev) 66 { 67 int i; 68 for (i = 0; i < VIRTIO_BALLOON_S_NR; dev->stats[i++] = -1); 69 } 70 71 static bool balloon_stats_supported(const VirtIOBalloon *s) 72 { 73 VirtIODevice *vdev = VIRTIO_DEVICE(s); 74 return virtio_vdev_has_feature(vdev, VIRTIO_BALLOON_F_STATS_VQ); 75 } 76 77 static bool balloon_stats_enabled(const VirtIOBalloon *s) 78 { 79 return s->stats_poll_interval > 0; 80 } 81 82 static void balloon_stats_destroy_timer(VirtIOBalloon *s) 83 { 84 if (balloon_stats_enabled(s)) { 85 timer_del(s->stats_timer); 86 timer_free(s->stats_timer); 87 s->stats_timer = NULL; 88 s->stats_poll_interval = 0; 89 } 90 } 91 92 static void balloon_stats_change_timer(VirtIOBalloon *s, int64_t secs) 93 { 94 timer_mod(s->stats_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + secs * 1000); 95 } 96 97 static void balloon_stats_poll_cb(void *opaque) 98 { 99 VirtIOBalloon *s = opaque; 100 VirtIODevice *vdev = VIRTIO_DEVICE(s); 101 102 if (s->stats_vq_elem == NULL || !balloon_stats_supported(s)) { 103 /* re-schedule */ 104 balloon_stats_change_timer(s, s->stats_poll_interval); 105 return; 106 } 107 108 virtqueue_push(s->svq, s->stats_vq_elem, s->stats_vq_offset); 109 virtio_notify(vdev, s->svq); 110 g_free(s->stats_vq_elem); 111 s->stats_vq_elem = NULL; 112 } 113 114 static void balloon_stats_get_all(Object *obj, Visitor *v, const char *name, 115 void *opaque, Error **errp) 116 { 117 Error *err = NULL; 118 VirtIOBalloon *s = opaque; 119 int i; 120 121 visit_start_struct(v, name, NULL, 0, &err); 122 if (err) { 123 goto out; 124 } 125 visit_type_int(v, "last-update", &s->stats_last_update, &err); 126 if (err) { 127 goto out_end; 128 } 129 130 visit_start_struct(v, "stats", NULL, 0, &err); 131 if (err) { 132 goto out_end; 133 } 134 for (i = 0; i < VIRTIO_BALLOON_S_NR; i++) { 135 visit_type_uint64(v, balloon_stat_names[i], &s->stats[i], &err); 136 if (err) { 137 goto out_nested; 138 } 139 } 140 visit_check_struct(v, &err); 141 out_nested: 142 visit_end_struct(v, NULL); 143 144 if (!err) { 145 visit_check_struct(v, &err); 146 } 147 out_end: 148 visit_end_struct(v, NULL); 149 out: 150 error_propagate(errp, err); 151 } 152 153 static void balloon_stats_get_poll_interval(Object *obj, Visitor *v, 154 const char *name, void *opaque, 155 Error **errp) 156 { 157 VirtIOBalloon *s = opaque; 158 visit_type_int(v, name, &s->stats_poll_interval, errp); 159 } 160 161 static void balloon_stats_set_poll_interval(Object *obj, Visitor *v, 162 const char *name, void *opaque, 163 Error **errp) 164 { 165 VirtIOBalloon *s = opaque; 166 Error *local_err = NULL; 167 int64_t value; 168 169 visit_type_int(v, name, &value, &local_err); 170 if (local_err) { 171 error_propagate(errp, local_err); 172 return; 173 } 174 175 if (value < 0) { 176 error_setg(errp, "timer value must be greater than zero"); 177 return; 178 } 179 180 if (value > UINT32_MAX) { 181 error_setg(errp, "timer value is too big"); 182 return; 183 } 184 185 if (value == s->stats_poll_interval) { 186 return; 187 } 188 189 if (value == 0) { 190 /* timer=0 disables the timer */ 191 balloon_stats_destroy_timer(s); 192 return; 193 } 194 195 if (balloon_stats_enabled(s)) { 196 /* timer interval change */ 197 s->stats_poll_interval = value; 198 balloon_stats_change_timer(s, value); 199 return; 200 } 201 202 /* create a new timer */ 203 g_assert(s->stats_timer == NULL); 204 s->stats_timer = timer_new_ms(QEMU_CLOCK_VIRTUAL, balloon_stats_poll_cb, s); 205 s->stats_poll_interval = value; 206 balloon_stats_change_timer(s, 0); 207 } 208 209 static void virtio_balloon_handle_output(VirtIODevice *vdev, VirtQueue *vq) 210 { 211 VirtIOBalloon *s = VIRTIO_BALLOON(vdev); 212 VirtQueueElement *elem; 213 MemoryRegionSection section; 214 215 for (;;) { 216 size_t offset = 0; 217 uint32_t pfn; 218 elem = virtqueue_pop(vq, sizeof(VirtQueueElement)); 219 if (!elem) { 220 return; 221 } 222 223 while (iov_to_buf(elem->out_sg, elem->out_num, offset, &pfn, 4) == 4) { 224 hwaddr pa; 225 hwaddr addr; 226 int p = virtio_ldl_p(vdev, &pfn); 227 228 pa = (hwaddr) p << VIRTIO_BALLOON_PFN_SHIFT; 229 offset += 4; 230 231 section = memory_region_find(get_system_memory(), pa, 232 BALLOON_PAGE_SIZE); 233 if (!section.mr) { 234 trace_virtio_balloon_bad_addr(pa); 235 continue; 236 } 237 if (!memory_region_is_ram(section.mr) || 238 memory_region_is_rom(section.mr) || 239 memory_region_is_romd(section.mr)) { 240 trace_virtio_balloon_bad_addr(pa); 241 memory_region_unref(section.mr); 242 continue; 243 } 244 245 trace_virtio_balloon_handle_output(memory_region_name(section.mr), 246 pa); 247 /* Using memory_region_get_ram_ptr is bending the rules a bit, but 248 should be OK because we only want a single page. */ 249 addr = section.offset_within_region; 250 balloon_page(memory_region_get_ram_ptr(section.mr) + addr, 251 !!(vq == s->dvq)); 252 memory_region_unref(section.mr); 253 } 254 255 virtqueue_push(vq, elem, offset); 256 virtio_notify(vdev, vq); 257 g_free(elem); 258 } 259 } 260 261 static void virtio_balloon_receive_stats(VirtIODevice *vdev, VirtQueue *vq) 262 { 263 VirtIOBalloon *s = VIRTIO_BALLOON(vdev); 264 VirtQueueElement *elem; 265 VirtIOBalloonStat stat; 266 size_t offset = 0; 267 qemu_timeval tv; 268 269 elem = virtqueue_pop(vq, sizeof(VirtQueueElement)); 270 if (!elem) { 271 goto out; 272 } 273 274 if (s->stats_vq_elem != NULL) { 275 /* This should never happen if the driver follows the spec. */ 276 virtqueue_push(vq, s->stats_vq_elem, 0); 277 virtio_notify(vdev, vq); 278 g_free(s->stats_vq_elem); 279 } 280 281 s->stats_vq_elem = elem; 282 283 /* Initialize the stats to get rid of any stale values. This is only 284 * needed to handle the case where a guest supports fewer stats than it 285 * used to (ie. it has booted into an old kernel). 286 */ 287 reset_stats(s); 288 289 while (iov_to_buf(elem->out_sg, elem->out_num, offset, &stat, sizeof(stat)) 290 == sizeof(stat)) { 291 uint16_t tag = virtio_tswap16(vdev, stat.tag); 292 uint64_t val = virtio_tswap64(vdev, stat.val); 293 294 offset += sizeof(stat); 295 if (tag < VIRTIO_BALLOON_S_NR) 296 s->stats[tag] = val; 297 } 298 s->stats_vq_offset = offset; 299 300 if (qemu_gettimeofday(&tv) < 0) { 301 warn_report("%s: failed to get time of day", __func__); 302 goto out; 303 } 304 305 s->stats_last_update = tv.tv_sec; 306 307 out: 308 if (balloon_stats_enabled(s)) { 309 balloon_stats_change_timer(s, s->stats_poll_interval); 310 } 311 } 312 313 static void virtio_balloon_get_config(VirtIODevice *vdev, uint8_t *config_data) 314 { 315 VirtIOBalloon *dev = VIRTIO_BALLOON(vdev); 316 struct virtio_balloon_config config = {}; 317 318 config.num_pages = cpu_to_le32(dev->num_pages); 319 config.actual = cpu_to_le32(dev->actual); 320 321 trace_virtio_balloon_get_config(config.num_pages, config.actual); 322 memcpy(config_data, &config, sizeof(struct virtio_balloon_config)); 323 } 324 325 static int build_dimm_list(Object *obj, void *opaque) 326 { 327 GSList **list = opaque; 328 329 if (object_dynamic_cast(obj, TYPE_PC_DIMM)) { 330 DeviceState *dev = DEVICE(obj); 331 if (dev->realized) { /* only realized DIMMs matter */ 332 *list = g_slist_prepend(*list, dev); 333 } 334 } 335 336 object_child_foreach(obj, build_dimm_list, opaque); 337 return 0; 338 } 339 340 static ram_addr_t get_current_ram_size(void) 341 { 342 GSList *list = NULL, *item; 343 ram_addr_t size = ram_size; 344 345 build_dimm_list(qdev_get_machine(), &list); 346 for (item = list; item; item = g_slist_next(item)) { 347 Object *obj = OBJECT(item->data); 348 if (!strcmp(object_get_typename(obj), TYPE_PC_DIMM)) { 349 size += object_property_get_int(obj, PC_DIMM_SIZE_PROP, 350 &error_abort); 351 } 352 } 353 g_slist_free(list); 354 355 return size; 356 } 357 358 static void virtio_balloon_set_config(VirtIODevice *vdev, 359 const uint8_t *config_data) 360 { 361 VirtIOBalloon *dev = VIRTIO_BALLOON(vdev); 362 struct virtio_balloon_config config; 363 uint32_t oldactual = dev->actual; 364 ram_addr_t vm_ram_size = get_current_ram_size(); 365 366 memcpy(&config, config_data, sizeof(struct virtio_balloon_config)); 367 dev->actual = le32_to_cpu(config.actual); 368 if (dev->actual != oldactual) { 369 qapi_event_send_balloon_change(vm_ram_size - 370 ((ram_addr_t) dev->actual << VIRTIO_BALLOON_PFN_SHIFT)); 371 } 372 trace_virtio_balloon_set_config(dev->actual, oldactual); 373 } 374 375 static uint64_t virtio_balloon_get_features(VirtIODevice *vdev, uint64_t f, 376 Error **errp) 377 { 378 VirtIOBalloon *dev = VIRTIO_BALLOON(vdev); 379 f |= dev->host_features; 380 virtio_add_feature(&f, VIRTIO_BALLOON_F_STATS_VQ); 381 return f; 382 } 383 384 static void virtio_balloon_stat(void *opaque, BalloonInfo *info) 385 { 386 VirtIOBalloon *dev = opaque; 387 info->actual = get_current_ram_size() - ((uint64_t) dev->actual << 388 VIRTIO_BALLOON_PFN_SHIFT); 389 } 390 391 static void virtio_balloon_to_target(void *opaque, ram_addr_t target) 392 { 393 VirtIOBalloon *dev = VIRTIO_BALLOON(opaque); 394 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 395 ram_addr_t vm_ram_size = get_current_ram_size(); 396 397 if (target > vm_ram_size) { 398 target = vm_ram_size; 399 } 400 if (target) { 401 dev->num_pages = (vm_ram_size - target) >> VIRTIO_BALLOON_PFN_SHIFT; 402 virtio_notify_config(vdev); 403 } 404 trace_virtio_balloon_to_target(target, dev->num_pages); 405 } 406 407 static int virtio_balloon_post_load_device(void *opaque, int version_id) 408 { 409 VirtIOBalloon *s = VIRTIO_BALLOON(opaque); 410 411 if (balloon_stats_enabled(s)) { 412 balloon_stats_change_timer(s, s->stats_poll_interval); 413 } 414 return 0; 415 } 416 417 static const VMStateDescription vmstate_virtio_balloon_device = { 418 .name = "virtio-balloon-device", 419 .version_id = 1, 420 .minimum_version_id = 1, 421 .post_load = virtio_balloon_post_load_device, 422 .fields = (VMStateField[]) { 423 VMSTATE_UINT32(num_pages, VirtIOBalloon), 424 VMSTATE_UINT32(actual, VirtIOBalloon), 425 VMSTATE_END_OF_LIST() 426 }, 427 }; 428 429 static void virtio_balloon_device_realize(DeviceState *dev, Error **errp) 430 { 431 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 432 VirtIOBalloon *s = VIRTIO_BALLOON(dev); 433 int ret; 434 435 virtio_init(vdev, "virtio-balloon", VIRTIO_ID_BALLOON, 436 sizeof(struct virtio_balloon_config)); 437 438 ret = qemu_add_balloon_handler(virtio_balloon_to_target, 439 virtio_balloon_stat, s); 440 441 if (ret < 0) { 442 error_setg(errp, "Only one balloon device is supported"); 443 virtio_cleanup(vdev); 444 return; 445 } 446 447 s->ivq = virtio_add_queue(vdev, 128, virtio_balloon_handle_output); 448 s->dvq = virtio_add_queue(vdev, 128, virtio_balloon_handle_output); 449 s->svq = virtio_add_queue(vdev, 128, virtio_balloon_receive_stats); 450 451 reset_stats(s); 452 } 453 454 static void virtio_balloon_device_unrealize(DeviceState *dev, Error **errp) 455 { 456 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 457 VirtIOBalloon *s = VIRTIO_BALLOON(dev); 458 459 balloon_stats_destroy_timer(s); 460 qemu_remove_balloon_handler(s); 461 virtio_cleanup(vdev); 462 } 463 464 static void virtio_balloon_device_reset(VirtIODevice *vdev) 465 { 466 VirtIOBalloon *s = VIRTIO_BALLOON(vdev); 467 468 if (s->stats_vq_elem != NULL) { 469 virtqueue_unpop(s->svq, s->stats_vq_elem, 0); 470 g_free(s->stats_vq_elem); 471 s->stats_vq_elem = NULL; 472 } 473 } 474 475 static void virtio_balloon_set_status(VirtIODevice *vdev, uint8_t status) 476 { 477 VirtIOBalloon *s = VIRTIO_BALLOON(vdev); 478 479 if (!s->stats_vq_elem && vdev->vm_running && 480 (status & VIRTIO_CONFIG_S_DRIVER_OK) && virtqueue_rewind(s->svq, 1)) { 481 /* poll stats queue for the element we have discarded when the VM 482 * was stopped */ 483 virtio_balloon_receive_stats(vdev, s->svq); 484 } 485 } 486 487 static void virtio_balloon_instance_init(Object *obj) 488 { 489 VirtIOBalloon *s = VIRTIO_BALLOON(obj); 490 491 object_property_add(obj, "guest-stats", "guest statistics", 492 balloon_stats_get_all, NULL, NULL, s, NULL); 493 494 object_property_add(obj, "guest-stats-polling-interval", "int", 495 balloon_stats_get_poll_interval, 496 balloon_stats_set_poll_interval, 497 NULL, s, NULL); 498 } 499 500 static const VMStateDescription vmstate_virtio_balloon = { 501 .name = "virtio-balloon", 502 .minimum_version_id = 1, 503 .version_id = 1, 504 .fields = (VMStateField[]) { 505 VMSTATE_VIRTIO_DEVICE, 506 VMSTATE_END_OF_LIST() 507 }, 508 }; 509 510 static Property virtio_balloon_properties[] = { 511 DEFINE_PROP_BIT("deflate-on-oom", VirtIOBalloon, host_features, 512 VIRTIO_BALLOON_F_DEFLATE_ON_OOM, false), 513 DEFINE_PROP_END_OF_LIST(), 514 }; 515 516 static void virtio_balloon_class_init(ObjectClass *klass, void *data) 517 { 518 DeviceClass *dc = DEVICE_CLASS(klass); 519 VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass); 520 521 dc->props = virtio_balloon_properties; 522 dc->vmsd = &vmstate_virtio_balloon; 523 set_bit(DEVICE_CATEGORY_MISC, dc->categories); 524 vdc->realize = virtio_balloon_device_realize; 525 vdc->unrealize = virtio_balloon_device_unrealize; 526 vdc->reset = virtio_balloon_device_reset; 527 vdc->get_config = virtio_balloon_get_config; 528 vdc->set_config = virtio_balloon_set_config; 529 vdc->get_features = virtio_balloon_get_features; 530 vdc->set_status = virtio_balloon_set_status; 531 vdc->vmsd = &vmstate_virtio_balloon_device; 532 } 533 534 static const TypeInfo virtio_balloon_info = { 535 .name = TYPE_VIRTIO_BALLOON, 536 .parent = TYPE_VIRTIO_DEVICE, 537 .instance_size = sizeof(VirtIOBalloon), 538 .instance_init = virtio_balloon_instance_init, 539 .class_init = virtio_balloon_class_init, 540 }; 541 542 static void virtio_register_types(void) 543 { 544 type_register_static(&virtio_balloon_info); 545 } 546 547 type_init(virtio_register_types) 548